Evolution in Value Relevance of Accounting Information

(For best results, please print in color)

Mary E. Bartha

Ken Li

Charles G. McClure

Graduate School of Business

Stanford Universityb

May 2018

We thank Susan Athey, Greg Clinch, Bob Holthausen, Guido Imbens, Bjorn Jorgensen, Jian

Kang, Lester Mackey, Thomas Scott, Eric So, Catalin Starica, Mary Tokar, Youfei Xiao, and

seminar participants at Cass Business School, ESSEC Business School, European Accounting

Association Annual Congress, Indiana University, London School of Economics, University of

Melbourne, New York University, University of Oxford, and Stanford University for helpful

comments and suggestions.

a Corresponding author. Email: mbarth@stanford.edu b 655 Knight Way, Stanford, CA 94305

Evolution in Value Relevance of Accounting Information

Abstract

We address how value relevance of accounting information evolved as the new economy

developed. Prior research concludes accounting information—primarily earnings—has lost

relevance. We consider more accounting amounts and find no decline in combined value

relevance from 1962 to 2014. We assess evolution in each amount’s value relevance and find

increases, most notably for amounts related to intangible assets, growth opportunities, and

alternative performance measures, which are important in the new economy. The number of

relevant amounts also increases. We also consider separately new economy, non-new economy

profit, and non-new economy loss firms. Although the relevance trends are most pronounced for

new economy firms, they are economy-wide. We base inferences on a non-parametric approach

that automatically incorporates nonlinearities and interactions, thereby unconstraining the

valuation relation. Taken together, our findings reveal a more nuanced, but not declining,

relation between share price and accounting information that reflects the new economy.

JEL classification: C14, G10, G18, M40, M41

Keywords: Capital Markets; Equity Valuation; Financial Reporting; Value Relevance; New

Economy; Classification and Regression Trees;

1

Evolution in Value Relevance of Accounting Information

1. Introduction

The question we address is how the value relevance of accounting information evolved as

the economy transitioned from primarily industrial to an economy based on services and

information technology. Prior research finds value relevance of accounting amounts—

particularly earnings—has declined, attributes the decline to the rise of this new economy, and

concludes accounting information has lost its relevance.1 We consider value relevance of a

larger set of accounting amounts, including amounts that could reflect information about

intangible assets, growth opportunities, and alternative firm performance measures, which are

important in the new economy. We find increases in relevance of these accounting amounts

combine to offset earnings’ relevance decline. We also find that although these trends are more

pronounced for firms emblematic of the new economy, they also are evident for other firms,

which suggests they are economy-wide. Our findings reveal an evolution in value relevance of

accounting information to a more nuanced, not declining, relation between accounting amounts

and share price that reflects equity valuation in the new economy.

Understanding the evolution in value relevance of accounting information provides

insights into whether accounting is relevant in the new economy. This is not a given because

accounting developed when the US economy largely comprised industrial firms and has not

changed fundamentally since then. It is well-known that current accounting does not include

intangible asset values, does not attempt to reflect growth opportunities, and emphasizes a single

performance measure, namely earnings. Determining the evolution in value relevance of

1 Consistent with prior research, we define an accounting amount as value relevant if it explains variation in share

price (Barth, Beaver, and Landsman 2001). We use the terms net income and earnings interchangeably.

2

individual accounting amounts reveals the extent to which each amount has become more and

less relevant. Thus, our findings provide insights into how accounting information reflects, and

might be enhanced to reflect better, information investors use when assessing firm value and

provide insights potentially relevant to equity valuation. In particular, finding increased

relevance of accounting amounts relating to intangible assets, growth opportunities, and

alternative performance measures—despite incomplete reflection in accounting—reveals these

categories of information are relevant to investors in the new economy.

We base our inferences on annual relations between equity price and accounting amounts

from 1962 to 2014, and measure value relevance as the explanatory power of the estimated

relation. Prior research largely estimates a linear relation, and does not identify precisely how

each accounting amount maps into future cash flows and therefore into equity value. Thus,

imposing a particular functional form could understate the accounting amounts’ explanatory

power and, consequently, their value relevance. In contrast, we employ a flexible, non-

parametric estimation approach—Classification and Regression Trees (CART)—that

automatically incorporates any nonlinearities and interactions, thereby permitting the accounting

amounts’ value relevance to manifest more fully.

We measure value relevance using out-of-sample explanatory power because in-sample

explanatory power can overstate value relevance. We find CART’s average out-of-sample

explanatory power across years is higher than that of a relation that is linear in the accounting

amounts and includes interactions with industry membership and whether the firm reports a

loss—69.3% versus 28.7%. Strikingly, the in-sample explanatory power of latter specification—

which is not uncommon in accounting research—averages 71.9%, which indicates it is

substantially overstated. We also find the out-of-sample explanatory power of a linear relation

3

that includes only earnings and equity book value averages 55.4%. The higher explanatory

power for CART—69.3% versus 55.4%—reflects additional explanatory power associated with

more accounting amounts and inclusion of nonlinearities and interactions.

The accounting amounts we consider are net income and equity book value, fourteen

other amounts presented in financial statements throughout our sample period that prior research

finds are value relevant, and ten industry indicators (hereafter, “accounting amounts”). Although

prior research assessing trends in value relevance of accounting information considers some of

these amounts, no study considers them all. Also, we are unaware of studies that assess trends in

value relevance of individual accounting amounts, other than earnings and equity book value.

Relating to intangible assets we consider research and development expense, recognized

intangible assets, and advertising expense, and relating to growth opportunities we consider cash

and revenue growth. Relating to alternative performance measures we consider operating cash

flow and revenue, and we consider special items and other comprehensive income because they

often are excluded when constructing non-GAAP performance measures and Street earnings.

We also consider dividends, capital expenditures, cost of goods sold (COGS), selling, general,

and administrative expense (SGA), and total assets. We include industry indicators to allow the

relation between price and the accounting amounts to vary across industries, which prior research

shows is important.

We first re-examine the conclusion of prior research that the value relevance of

accounting information has declined. We do not expect a decline because we consider more

accounting amounts and employ a flexible, non-parametric estimation method. Consistent with

this expectation, we find no evidence of a decline in value relevance of accounting information

across all sample years or in any decade other than the 1990s, which coincides with the

4

technology bubble. If anything, we find some evidence of an increase. Thus, in contrast to the

overall conclusion from prior research, our findings do not support a conclusion that the value

relevance of accounting information has declined.

We next determine how much of combined value relevance derives from each accounting

amount. This determination permits us to provide insights into how individual amounts

contribute to the evolution in value relevance of accounting information. Based on prior

research, we expect the relevance of earnings to decline and that of equity book value to

increase. We find that they do. More importantly for our research question, we expect

accounting amounts related to intangible assets, growth opportunities, and alternative

performance measures to become more relevant. Consistent with these expectations, we find

significant increases in relevance of research and development expense, recognized intangible

assets, cash, revenue growth, operating cash flow, revenue, and special items.

Because prior research finds a decline in the number of dividend-paying firms, we expect

the relevance of dividends to decline. We find that it does. Although we have no expectations

regarding trends in relevance of the remaining accounting amounts, we find a significant decline

(increase) in relevance of COGS and total assets (capital expenditure). We also find a significant

increase in the number of accounting amounts needed to explain the same proportion of

combined value relevance, which indicates more accounting amounts reflect information in price

and suggests valuations have become more nuanced.

We next examine the evolution in value relevance of individual accounting amounts

separately for New Economy firms, Non-New Economy Profit firms, and Non-New Economy

Loss firms. New Economy firms are those with at least two characteristics prior research

identifies as emblematic of firms in the new economy, namely being young, operating in a

5

technology industry, and reporting a loss. We separate Non-New Economy firms into those

reporting a profit and those reporting a loss because prior research shows accounting amounts—

particularly earnings—of profit and loss firms have different relations with price, and loss firms

are notoriously difficult to value. Over the last five decades the economy proportions of New

Economy firms and Non-New Economy Loss firms significantly increased. Although the

proportion of Non-New Economy Profit firms significantly decreased, these firms represent the

largest proportion of firms in the economy over the sample period.

Regarding value relevance of earnings, we find its relevance has declined for New

Economy firms and Non-New Economy Profit firms. As expected, we find the decline is more

dramatic for New Economy firms. Not surprisingly, we find earnings has little relevance for

Non-New Economy Loss firms. Thus, the increase in proportion of New Economy and Non-

New Economy Loss firms contributes to earnings’ value relevance decline in the full sample.

Regarding value relevance of accounting amounts relating to intangible assets, growth

opportunities, and alternative performance measures, as expected, we find significant increases

for New Economy firms and that the increases relating to intangible assets and growth

opportunities are most pronounced for these firms. However, we also find the relevance of these

accounting amounts—and those related to alternative performance measures—also increase for

Non-New Economy firms, and generally significantly so. These findings suggest firms

emblematic of the new economy are not solely responsible for the relevance increases. Rather,

they reflect economy-wide changes.

The remainder of the paper proceeds as follows. Section 2 relates our study to prior

research, and describes the accounting amounts and our predictions. Section 3 develops the

6

research design. Section 4 describes the sample and data and section 5 presents the findings.

Section 6 presents additional analyses and section 7 concludes.

2. Related literature, accounting amounts, and predictions

2.1 Research on trends in value relevance of accounting amounts

A large literature documents a decline in earnings’ value relevance and offers two

primary explanations for the decline. The first is the rise of the new economy in which future

earnings largely depends on investments in intangible assets. Investments in intangible assets

typically are expensed as incurred even though the investments generate economic benefits over

a longer period, which negatively affects earnings quality. Lev and Zarowin (1999) finds a

weaker association between price and earnings for firms with more intangible assets and

attributes this finding to the timing mismatch of expenses and revenue associated with such

assets. Consistent with a decrease in matching, Dichev and Tang (2008) finds a decrease in the

correlation between revenue and expenses. Donelson, Jennings, and McInnis (2011) and

Srivastava (2014) suggest the decline in matching is attributable to changes in the economy, such

as an increase in the number of new firms with business models focused on intangible assets.2

The second explanation is the presence of more loss firms. Hayn (1995) and Collins, Pincus, and

Xie (1999) find earnings is less relevant for loss firms, and Barth, Beaver, and Landsman (1998)

finds the relevance of net income (equity book value) decreases (increases) as a firm’s financial

health decreases. Thus, Collins et al. (1999) suggests the presence of more loss firms explain

earnings’ relevance decline.3

2 Dichev and Tang (2008) explains matching could have declined because of changes in the economy or changes in

accounting standards. As is Dichev and Tang (2008), we are interested in the combined effect of these changes

because both influence the extent to which accounting amounts reflect information investors use to value firms. 3 A third explanation is increased noise in equity prices. Brunnermeier and Nagel (2004) finds hedge funds traded

against fundamentals from 1998 to 2000, which was the height of the technology bubble, and Dontoh,

Radhakrishnan, and Ronen (2004) posits this increased presence of noise traders exacerbates the decline in value

7

Collins, Maydew, and Weiss (1997) suggests that although shifts in the economy towards

technology and loss firms could explain the decline in earnings’ relevance, these shifts could

increase equity book value’s relevance. This could occur because equity book value predicts

future normal earnings and reflects loss firms’ abandonment option (Barth et al. 1998a; Collins

et al. 1999). Consistent with this logic, Collins et al. (1997) and Francis and Schipper (1999)

find earnings’ value relevance decline is offset by an increase in equity book value’s relevance.

Brown, Lo, and Lys (1999) also finds a(n) decline (increase) in value relevance of earnings

(equity book value) from the late 1950s to the late 1990s, but finds a decline in their combined

relevance after including controls for scale effects. Lev and Zarowin (1999) and Balachandran

and Mohanram (2011) find a decline in combined value relevance of earnings and equity book

value from the late 1970s to the early 2000s.

Studies considering accounting amounts in addition to earnings and equity book value

also find a decline in value relevance of accounting information. Core, Guay, and Van Buskirk

(2003) considers research and development expense, advertising expense, capital expenditure,

and revenue growth in addition to earnings and equity book value, and finds their combined

value relevance is lower from 1996 to 1999 than from 1975 to 1995. Core et al. (2003) explains

that revenue growth is included because valuation of high technology firms is influenced by

future growth opportunities to a greater extent than other firms. Lev and Gu (2016) considers

earnings and assets and liabilities—thereby, indirectly, equity book value—and selected

components of earnings, namely revenue, COGS, and SGA, and finds a decline in their

relevance of accounting information because such traders do not trade based on fundamental information.

Consistent with this explanation, Core, Guay, and Van Buskirk (2003) finds the value relevance of the accounting

amounts that study considers is lower from 1996 to 1999 than in prior periods, but the accounting amounts’

coefficients are stable. Core et al. (2003) interprets these findings as indicating the lower value relevance is

attributable to increases in price and returns variances unrelated to accounting information.

8

combined value relevance from 1950 to 2013. Based on this evidence, Lev and Gu (2016)

proclaims the “end of accounting.” However, other studies find accounting amounts these

studies do not consider also are value relevant (Barth et al. 2001; Holthausen and Watts 2001).

Thus, the overall inference of a decline in relevance of accounting information could result from

failure to consider these amounts.4

We contribute to this literature in four primary ways. First, we assess evolution in

combined value relevance of accounting information using more accounting amounts, including

some the reflect information relevant in the new economy. Second, we assess evolution in value

relevance for each accounting amount to provide direct evidence on which amounts have

increased and declined in relevance. Third, we assess evolution in value relevance of each

accounting amount separately for new economy, non-new economy profit, and non-new

economy loss firms to determine the extent to which trends in relevance are economy-wide.

Fourth, we measure value relevance as the out-of-sample explanatory power of the relation

between price and accounting amounts estimated using a flexible, non-parametric approach that

automatically incorporates nonlinearities and interactions, thereby permitting the accounting

amounts’ value relevance to manifest more fully.

2.2 Accounting amounts and predictions

This section identifies the accounting amounts we consider, explains why we consider

them, and provides our expectations as to trends in their value relevance. In brief, we consider

amounts from prior research on trends in value relevance of accounting information, as well as

amounts presented in financial statements throughout our sample period that prior research finds

4 Davern, Gyles, Hanlon, and Pinnuck (2018) finds no decline in value relevance of earnings and equity book value

for Australian firms from 1992 to 2015, but finds EBIT and EBITDA are more value relevant than earnings. From

1996 to 2014, Filip, Ghio, and Paugam (2018) finds earnings (cash flow) relevance is lower (higher) for innovative

Small and Medium Entities (SMEs) listed on the AIM London Stock exchange than for non-innovative SMEs.

9

are value relevant. These include net income and equity book value, fourteen other amounts

presented in financial statements, and ten industry indicators.

Before assessing trends in individual accounting amounts, we re-examine the conclusion

in prior research that the value relevance of accounting information has declined. We do not

expect to find a decline because we consider more accounting amounts, some of which could

reflect information relevant in the new economy, and we employ a flexible, non-parametric

estimation method. However, our expectations may not be borne out because current accounting

does not include intangible asset values, does not attempt to reflect growth opportunities, and

emphasizes a single performance measure, namely earnings. Thus, accounting amounts might

not reflect information relevant to valuing firms in the new economy.

2.2.1 Net income and equity book value

Following a large literature on value relevance of accounting, we consider net income

and equity book value, the two primary accounting summary measures. Miller and Modigliani

(1966) and Ohlson (1995) theoretically support considering these amounts. Based on the prior

research in section 2.1, we expect earnings’ (equity book value’s) relevance to decline (increase).

2.2.2 Intangible assets

Regarding intangible assets, we consider research and development expense and

advertising expense because Core et al. (2003) suggests these accounting amounts reflect

expected future earnings growth associated with investments in intangible assets. Lev and

Sougiannis (1996) finds research and development and advertising expenses have positive

relations with future operating earnings because these expenses indicate investments in

technological innovation and brand awareness. Barth, Clement, Foster, and Kasznik (1998) finds

advertising expense is associated with brand values, which are value relevant. We also consider

10

recognized intangible assets. Aboody and Lev (1998) finds capitalized software development

costs—a type of recognized intangible asset—is associated with future earnings, which suggests

recognized intangible assets are value relevant. We expect these accounting amounts to become

more relevant with the rise of the new economy in which intangible assets are important to firm

value. However, our expectations might not be borne out because these accounting amounts are

not designed to capture intangible asset values, e.g., advertising expense is not designed to

capture brand value.

2.2.3 Growth opportunities

Regarding amounts that reflect information about growth opportunities, we consider cash

and revenue growth. Myers and Majluf (1984) suggests cash is more valuable when external

financing is costly and the firm has positive net present value projects. Opler, Pinkowitz, Stulz,

and Williamson (1999) finds firms with more growth opportunities hold more cash, which

suggests cash is more relevant for growth firms. Faulkender and Wang (2006) finds the marginal

value of cash is higher for firms with valuable investment opportunities and financial constraints.

Revenue growth captures expected growth in earnings (Core et al. 2003). We expect these

accounting amounts become more relevant with the rise of the new economy in which growth

opportunities are important to firm value. However, our expectations might not be borne out

because accounting amounts are not designed to capture, and thus likely incompletely reflect,

growth opportunities.

2.2.4 Alternative performance measures

Regarding performance measure alternatives to earnings, we consider operating cash

flow because it is an alternative to earnings when predicting future cash flows (e.g., Palepu and

Healy 2008). Operating cash flow also can be more persistent than earnings and predict future

11

earnings incremental to current earnings (Sloan 1996; Barth, Beaver, Hand, and Landsman

1999). Also, including operating cash flow together with earnings effectively includes accruals.

We consider revenue because it is particularly relevant for internet and loss firms (Davis 2002;

Callen, Robb, and Segal 2004), and Ertimur, Livnat, and Martikainen (2003) finds revenue is

more persistent than expenses.

We consider two other earnings components prior literature suggests have implications

for future cash flows that differ from earnings. The first is special items, which are less

persistent than other earnings components. Jones and Smith (2011) finds special items help

predict future net income and operating cash flow incremental to current net income. Collins et

al. (1997) documents that special items have increased, and suggests this increase could reduce

the value relevance of earnings. In addition, Bradshaw and Sloan (2002) and Bhattacharya,

Black, Christensen, and Larson (2003) find there is an increase in firms’ use of non-GAAP

earnings, which excludes unusual or non-recurring items, i.e., special items. Bradshaw and

Sloan (2002) also documents an increase in analyst Street earnings, which also generally

excludes special items. Including special items effectively includes earnings before special

items, which is substantially similar to operating earnings in Bhattacharya et al. (2003).

The second is other comprehensive income (OCI) because, as with special items, Jones

and Smith (2011) finds OCI has predictive power for future net income and operating cash flow

incremental to earnings. Jones and Smith (2011) observes OCI recognition of gains and losses

has increased, and firms and analysts often ignore OCI when constructing non-GAAP and Street

12

earnings. Including OCI in the accounting amounts effectively includes comprehensive income

because comprehensive income is the sum of net income and OCI.5

We expect an increase in value relevance of accounting amounts relating to alternative

performance measures for the reasons outlined above and given the decline in earnings quality

documented in prior research discussed in section 2.1. Our expectations might not be borne out

because, incremental to earnings, these accounting amounts might not reflect the performance

measures investors use when valuing the firm.

2.2.5 Other amounts

We also consider dividends, capital expenditure, COGS, SGA, total assets, and industry

indicators. Miller and Rock (1985) suggests dividends can be a signal about, and thus help

predict, future earnings in the presence of information asymmetry, thereby making dividends

relevant to firm value. Consistent with this suggestion, Watts (1973) finds a positive association

between dividends and future earnings. In addition, financial statement analysis textbooks

suggest dividend discount and dividend growth models can be appropriate for estimating firm

value (Palepu and Healy 2008). However, Floyd, Li, and Skinner (2015) finds a decrease from

1980 to 2012 in the propensity for firms, other than banks, to pay dividends, which leads us to

expect dividends’ relevance to decline. However, our expectations might not be borne out if

firms’ decreased propensity to pay dividends increases their relevance for firms that pay them.

Capital expenditure reflects investments in tangible assets. If tangible assets become

more important for firm value, capital expenditure could become more relevant. However, if the

new economy’s focus is only on intangible assets, capital expenditure could become less

5 Dhaliwal, Subramanyam, and Trezevant (1999) finds OCI has higher value relevance than net income for some

segments of the economy, such as financial firms, largely because OCI includes unrealized gains and losses on

marketable securities. As section 4 explains, we exclude financial firms from our main sample. However, section

6.2 describes untabulated findings for financial firms.

13

relevant. COGS and SGA can have implications for future earnings different from current

earnings (Lev and Gu 2016). Including assets effectively includes liabilities because liabilities is

the difference between assets and equity book value. Also, as section 2 explains, when using

CART including assets effectively includes asset-based ratios, such as return-on-assets and sales-

to-assets ratios, which prior research finds are relevant (Ou and Penman 1989). However, we

have no expectations regarding relevance trends for capital expenditure, COGS, SGA, or assets.

We include industry indicators because how accounting amounts map into firm value

differs for firms in different industries. Thus, it is common in value relevance research to

estimate valuation equations separately by industry (Barth et al. 1999). Also, as section 3

explains, when using CART, including industry indicators not only permits the relations between

the accounting amounts and price to vary across industries, but also enables the relations to

reflect interactions and nonlinearities between industry membership and the accounting amounts.

We have no expectation regarding value relevance of the industry indicators. If heterogeneity in

the relations between price and accounting amounts across industries increases, industry

indicators could become more relevant. If heterogeneity does not increase or other accounting

amounts, such as research and development expense, better capture the increase, the indicators

could become less relevant.

3. Research design

3.1 Estimating value relevance

To assess the evolution in value relevance of accounting information, we estimate, by

year, the relation between price and accounting amounts specified equation (1), and examine the

trend over time in its explanatory power.

𝑃𝑖 = 𝐶𝐴𝑅𝑇(𝑉𝐴𝑅𝑖 , 𝐼𝑁𝐷10𝑖) (1)

14

𝑃 is share price, 𝑉𝐴𝑅 is a vector of sixteen accounting amounts other than industry indicators,

and 𝐼𝑁𝐷10 is a set of indicator variables for the Fama-French ten industry groups. 𝑉𝐴𝑅

comprises earnings, 𝑁𝐼; equity book value, 𝐵𝑉𝐸; research and development expense, 𝑅𝐷;

recognized intangible assets, including capitalized software, goodwill, and other purchased

intangible assets, 𝐼𝑁𝑇𝐴𝑁; advertising expense, 𝐴𝐷𝑉; cash and short-term investments, 𝐶𝐴𝑆𝐻;

one-year revenue growth, 𝑅𝐸𝑉𝐺𝑅; operating cash flow, 𝐶𝐹; revenue, 𝑅𝐸𝑉; special items, 𝑆𝑃𝐼;

other comprehensive income, 𝑂𝐶𝐼; dividends to common shareholders, 𝐷𝐼𝑉; capital

expenditures, 𝐶𝐴𝑃𝑋; cost of goods sold, 𝐶𝑂𝐺𝑆; selling, general, and administrative expense,

𝑆𝐺𝐴; and total assets, 𝐴𝑆𝑆𝐸𝑇𝑆. See Appendix 1 for variable definitions. We measure price

three months after fiscal year-end to ensure the accounting information is publicly available.6

We deflate non-indicator amounts by shares outstanding to facilitate comparison with related

prior research (Collins et al. 1997; Brown et al. 1999; Lev and Zarowin 1999; Balachandran and

Mohanram 2011). Subscript i indexes firms.

𝐶𝐴𝑅𝑇 is the Classification and Regression Trees (CART) estimation function, which is

akin to a decision tree.7 Appendix 2 explains CART estimation. In brief, CART is a non-

parametric estimation approach that does not require the researcher to specify—and, thus

potentially inappropriately constrain—the relation’s functional form. CART identifies

nonlinearities in the underlying relation and interactions between and among the explanatory

6 Although investors can obtain information about firm value from financial statements or from other sources, our

value relevance metric reflects only the extent to which the accounting amounts, and information correlated with

them, explain price (Barth, Konchitchki, and Landsman 2013). 7 CART is part of a family of decision tree-based estimation methods that can increase explanatory power relative to

traditional linear methods, detect relative importance of explanatory variables, and uncover nonlinearities and

interactions in the marginal relations between the independent and dependent variables (Breiman 2001). These

methods are used in a variety of non-accounting research settings, including facial recognition, spam e-mail

detection, and disease diagnosis (see Verikas, Gelzinis, and Bacauskiene 2011 for a review), and in accounting

research (Gerakos, Hahn, Kovrijynkh, and Zhou 2016; Beaver, Cascino, Correia, and McNichols 2017; Correia,

Kang, and Richardson 2017; Jones 2017).

15

variables by recursively partitioning the variables, and automatically incorporates nonlinearities

and interactions that increase the relation’s out-of-sample explanatory power. These

nonlinearities and interactions effectively permit the relation between any explanatory variable

and price to vary as an unconstrained function of any of the other variables and to include any

ratios based on the explanatory variables. The absence of researcher-imposed constraints on the

relation is important because with potential nonlinearities and interactions in the underlying

relation, estimation that assumes an additively linear structure in the explanatory variables,

except for particular nonlinearities and interactions the researcher specifies, such as OLS, is

susceptible to understating value relevance (Hastie, Tibshirani, and Friedman 2001).

Even though researchers can specify known nonlinearities and interactions, more are

possible (Holthausen and Watts 2001). For example, Riffe and Thomson (1998) shows the

relation between price and earnings (equity book value) can be nonlinear when earnings (equity

book value) captures economic earnings (equity book value) with error. Yee (2000) shows a

nonlinear relation between price and earnings can exist when firms adapt to their competitive

environment. However, Riffe and Thompson (1998) and Yee (2000) observe that unmodeled

nonlinearities likely exist.8

Our value relevance metric is the out-of-sample R2 (OOS R2) from equation (1). We rely

on OOS R2 because when there are many explanatory variables relative to the number of

observations, in-sample adjusted R2, which is a commonly used value relevance metric, is

susceptible to overstating value relevance (Hastie et al. 2001). This occurs because with many

8 As section 2.1 explains, Collins et al. (1999) finds a flat relation between earnings and price for loss firms, but a

positive relation for profit firms. Barth et al. (1998a) finds valuation coefficients on earnings and equity book value

differ as a function of firms’ financial health. Faulkender and Wang (2006) finds firms with high growth

opportunities have a higher marginal value of cash. Also, a substantial literature examines the explanatory power of

various financial ratios, which are nonlinear transformations of accounting amounts (Ou and Penman 1989).

16

explanatory variables the estimation reflects sample characteristics rather than the underlying

relation. Thus, removing a few observations, re-estimating the equation using the remaining

observations, and using the resulting coefficient estimates to predict price for the removed

observations would result in poor predictive power. OOS R2 reveals such overstatement.

CART avoids overstating value relevance by bootstrapping the sample, estimating a tree

over each bootstrapped sample, using the tree from each bootstrapped sample to predict price for

out-of-sample observations, and comparing out of-sample price predictions—averaged across

bootstrapped samples—with actual price to estimate explanatory power. Fitting the equation

using a portion of the observations in each bootstrapped estimation and averaging estimates of

predicted price reduces noise in the estimated underlying relation.9 Because CART is estimated

over bootstrapped samples, in-sample R2 for CART is not well-defined.

To provide evidence that CART’s flexibility enables the accounting amounts’ value

relevance to manifest more fully, Appendix 2 reveals CART has higher mean OOS R2 across

sample years—69.3%—than two specifications commonly used in prior research. Mean out-of-

sample explanatory power is only 28.7% for OLS estimation of a specification that is linear in

the accounting amounts and includes interactions with industry and loss indicators. Mean in-

sample explanatory power for that specification is 71.9%, which indicates in-sample explanatory

power is substantially overstated. Mean out-of-sample explanatory power is 55.4% for OLS

estimation of a linear relation between price and earnings and equity book value. Its mean in-

sample explanatory power is 55.5%, which indicates less overstatement of explanatory power

9 CART estimation provides insights into value relevance trends, but has several limitations. First, although CART

automatically determines nonlinearities and interactions, it is difficult to visualize the estimated prediction function.

Thus, the tabulated findings and figures do not summarize all significant relations underlying the CART prediction

function. Second, using CART precludes us from summarizing findings in the form of coefficient estimates and

confidence intervals. Appendix 2 provides examples of nonlinearities in equation (1) that CART identifies.

17

with only two accounting amounts. The 13.9% (63.9% – 55.4%) higher OOS R2 for CART for

this specification reflects the additional explanatory power associated with more accounting

amounts and the inclusion of unspecified nonlinearities and interactions.10

3.2 Evolution in combined value relevance of accounting amounts

To test for a trend in combined value relevance, we estimate equation (2).

𝑂𝑂𝑆𝑅2𝑡 = 𝛽0 + 𝛽1𝑌𝐸𝐴𝑅𝑡 + 𝜀𝑡 (2)

𝑂𝑂𝑆𝑅2 is OOS R2 estimated from equation (1), and 𝑌𝐸𝐴𝑅 is year of the observation, 1962,…,

2014. t denotes year. Because we expect no decline in relevance of accounting information, we

expect 𝛽1 is positive or insignificantly different from zero.

3.3 Evolution in value relevance of each accounting amount

We next disaggregate each annual OOS R2 to determine how much value relevance

derives from each accounting amount. We test whether an accounting amount’s relevance has

increased or declined by estimating equation (3).

𝑉𝑅𝑘𝑡 = 𝛽0 + 𝛽1𝑌𝐸𝐴𝑅𝑡 + 𝜀𝑡 (3)

𝑉𝑅𝑘 is accounting amount 𝑘’s value relevance. When we expect an accounting amount has

become more (less) relevant, we expect 𝛽1 is positive (negative).

𝑉𝑅𝑘 is the percentage of OOS R2 attributable to accounting amount 𝑘. To construct 𝑉𝑅𝑘,

we determine how much combined value relevance decreases when we randomly assign each

accounting amount one at a time, relative to the other amounts.11 For example, if we randomly

10 To validate that CART does not spuriously generate high OOS R2, we estimate equation (1) for each year from

1962 to 2014 using prices selected randomly, with replacement, from the sample of actual prices for that year. The

untabulated findings reveal the OOS R2 is negative in every year, and the across-year mean is –6.0%. 11 Random assignment is equivalent to omitting the amount from the estimation, because it results in the amount

having no systematic relation with price or any other accounting amount. Thus, 𝑉𝑅𝑘 is analogous to the incremental R2 of variable k. However, in CART estimation, VR reflects the myriad ways in which an accounting amount is

value relevant, e.g., as a main effect, an interaction effect, and a nonlinearity effect, including its role in ratios.

18

assign 𝑁𝐼 and the resulting OOS R2 is less than if we randomly assign 𝐵𝑉𝐸, then 𝑁𝐼 is more

value relevant than 𝐵𝑉𝐸.12 We also estimate equation (3) using the sum of VRk for amounts

related to intangible assets, Intans—RD, INTAN, and ADV; growth opportunities, Growth—

CASH and REVGR; and alternative performance measures, AltPerf—CF, REV, SPI, and OCI.

To test for trends in the number of value relevant amounts, we estimate equation (4):

𝑁𝑈𝑀𝑉𝑅𝑡 = 𝛽0 + 𝛽1𝑌𝐸𝐴𝑅𝑡 + 𝜀𝑡. (4)

𝑁𝑈𝑀𝑉𝑅 is the number of relevant accounting amounts in year 𝑡 based on a specified relevance

threshold. In particular, we order the amounts by VR and, beginning with the amount with the

highest VR, we add amounts until they together explain 50%, 75%, 80%, 90%, and 95% of

combined value relevance.13 We interpret positive 𝛽1 as an increase in the nuance with which

price reflects accounting information.

3.4 Evolution in value relevance of accounting amounts by firm groups

We estimate the accounting amounts’ relevance separately for three groups of firms to

provide insights into the extent to which the evolution in combined value relevance applies only

to firms emblematic of the new economy or is economy-wide. For these estimations, the sample

begins in 1971 to ensure a sufficient sample size in each year and group.

The first group is New Economy firms. Prior research identifies technology industry,

loss, and young firms as emblematic of the new economy (Collins et al. 1997; Francis and

Schipper 1999; Core et al. 2003). Thus, we classify a firm as a New Economy firm if it has at

12 As an example, assume the combined explanatory power of NI and BVE is 0.90. Also assume that if NI (BVE) is

randomly assigned, the explanatory power is 0.30 (0.70), which means if NI (BVE) is randomly assigned the

explanatory power decreases by 0.60 (0.20) [0.90 – 0.30 (0.90 – 0.70)]. Thus, the value relevance of NI is 0.75

[0.60/(0.60 + 0.20)], and that of BVE is 0.25 [0.20/(0.60 + 0.20)]. We randomly assign one variable at a time to

assess its value relevance, except for the industry indicators, which we randomly assign together. 13 There is no standardized threshold for tests of value relevance of accounting amounts; we use these five.

19

least two of these characteristics.14 The second and third groups are Non-New Economy Profit

and Non-New Economy Loss firms. We distinguish profit and loss firms because prior research

finds earnings has higher (lower) relevance for profit (loss) firms. In addition, loss firms are

notoriously difficult to value (Joos and Plesko 2005; Darrough and Ye 2007). We provide

insights into which accounting amounts reflect investors’ assessments of value for these firms.15

4. Sample and descriptive statistics

The sample comprises 192,463 firm-year observations from Compustat and CRSP from

1962 to 2014. We begin the sample in 1962 when Compustat began its service (Fama and

French 1992; 1993; 2015). Although Compustat backfilled data prior to 1962, those data reflect

selection bias and coverage is tilted towards successful firms (Fama and French 1992;

Linnainmaa and Roberts 2016). Beginning in 1962 also enables us to detect trends associated

with the shift to the new economy, which is evident by the 1970s (Buera and Kaboski 2012). We

end the sample in 2014. Thus, the sample period covers a variety of economic conditions,

including the 2007-2008 financial crisis and the technology bubble, which we find is associated

with a drop and subsequent recovery in combined value relevance.

We require that firms list on NYSE, NASDAQ, or AMEX, are non-financial, and have

non-missing net income, equity book value, share price, number of shares outstanding, total

14 Technology firms are those in three-digit SIC industries with large unrecognized intangible assets, specifically

industries 283, 357, 360-368, 481, 737, and 873 (Francis and Schipper 1999; Core et al. 2003), which include

computer hardware and software, pharmaceuticals, electronic equipment, and telecommunications. Young firms

have been listed for five or fewer years (Core et al. 2003). Loss firms have negative earnings. 15 Although considering these groups separately provides insights into the evolution in value relevance of accounting amounts, the full sample findings are not simply a weighted average of the group findings. This is because CART

estimation uses all available observations. For example, NI could be more relevant in the full sample than as

indicated by the separate estimations for Non-New Economy Profit and Loss firms because in the full sample NI

distinguishes profit and loss firms, which is not necessary in the separate estimations.

20

assets, lagged total assets, revenue, and SIC code.16 If operating cash flow is missing, we

estimate it using the Sloan (1996) approach.17 We set to zero any other missing accounting

amount. We winsorize all non-indicator variables at the 1st and 99th percentiles, by year, to

mitigate the effect of outliers on estimation results.

Table 1 presents descriptive statistics and correlations for price and the non-indicator

accounting amounts. Panel A reveal mean price, P (net income, NI), is 18.14 (0.79), which is

higher (lower) than the 16.98 and 17.58 (1.10 and 1.29) in Brown et al. (1999) and Collins et al.

(1997). These differences from prior research reflect an increase (decrease) in price (earnings)

since the time of those studies, which is consistent with a decline in earnings’ relevance.

Table 1, panel B, presents correlations between the variables. We present Pearson

correlations to facilitate comparison to prior research. CART estimation is non-parametric and,

thus, any skewness in the distributions apparent from differences between Pearson and Spearman

correlations has no effect on our estimations. Regarding correlations with price, equity book

value has the largest, followed closely by net income. The Pearson (Spearman) correlations are

0.62 (0.74) for equity book value and 0.60 (0.70) for net income. Research and development

expense, which we code as positive, has a positive correlation with price (Pearson and Spearman

corrs. = 0.27 and 0.09), which is consistent with the expense reflecting investment in technology.

5. Findings

5.1 Value relevance over time

Table 2 presents findings relating to combined value relevance. Table 2 reveals mean

OOS R2 is 69.3% from 1962 to 2014. More importantly, table 2 reveals OOS R2 increases at a

16 We require lagged assets to ensure prior year accounting amounts are in Compustat because several variables,

such as other comprehensive income and revenue growth, require prior year amounts to compute. 17 Sloan (1996) estimates accruals by taking differences in changes in current assets and current liabilities. We

estimate operating cash flow as net income less estimated accruals.

21

rate of 0.218 percentage points per year (t-stat. = 2.88), which is inconsistent with a decline in

value relevance of accounting amounts.

Table 2 also reveals combined value relevance fluctuates across decades, but there is no

systematic decline. In particular, in the 1960s combined value relevance is 57.6% with no

significant within-decade trend (t-stat. = –0.31). In the 1970s it increases to 67.1%, with a

significant positive within-decade trend (t-stat. = 2.82). In the 1980s and 1990s, it increases to

78.1% and then decreases to 69.2%; the within-decade the trends are significantly positive and

negative (t-stats. = 4.03 and –4.67). Lower relevance in the 1990s and a declining within-decade

trend are consistent with investors trading away from fundamentals at the height of the

technology bubble (see footnote 3). They also can explain why prior research using sample

periods ending near the end of the 1990s finds a declining trend in combined value relevance

(e.g., Brown et al. 1999). In the 2000s and 2010s, combined value relevance is 70.8% and

71.7%, both with increasing within-decade trends (t-stats. = 2.14 and 2.85). Thus, compared to

the 1960s and 1970s, relevance of accounting information is higher in the 2000s and 2010s.18

Taken together, the findings in table 2 are consistent with our expectation of no significant

decline in the relevance of accounting information.

5.2 Value relevance of individual accounting amounts

Figure 1 shows the evolution in value relevance, VR, of each accounting amount, which

largely are consistent with our expectations. First, figure 1 reveals the decline (increase) in

relevance of NI (BVE). Second, it reveals an increase in number of relevant accounting amounts,

18 As section 2.2 explains, the accounting amounts are those available for the entire sample period. Yet, prior

research finds more recently available accounting amounts are value relevant, such as share repurchases and

issuances, total assets and total liabilities at fair value, stock-based compensation expense, and several pension

amounts (Barth et al. 2001; Aboody, Barth, and Kasznik 2004; Hribar, Jenkins, and Johnson 2006). Untabulated

findings from estimating equation (1) for 2014 reveal the CART OOS R2 is 73.8% (72.5%) with these variables

included (excluded), which suggests our combined value relevance estimates are lower than they would be if we

included more accounting amounts.

22

particularly those related to intangible assets, growth opportunities, and alternative performance

measures, which were barely noticeable in the 1960s. The pie charts display the accounting

amounts with the largest VR, in clockwise order, needed explain 90% of combined value

relevance (hereafter, important amounts). The pie charts reveal a stark contrast between the

1960s and 2010s. In the 1960s, the relevance of NI dominated that of other accounting amounts,

and there are only five important amounts, none of which relate to intangible assets, growth

opportunities, or alternative performance measures. In the 2010s, NI is considerably less

dominant, and there are eight other important amounts, five of which relate to intangible assets,

growth opportunities, and alternative performance measures.

The sharp decline in earnings’ relevance in the late 1990s and early 2000s largely is

associated with the technology bubble.19 Untabulated statistics reveal that in 1999, the year of

lowest relevance of earnings, cash and revenue growth are the most important accounting

amounts. This finding reveals growth opportunities were more relevant than earnings at the

height of the technology bubble. Figure 1 reveals that subsequently the relevance of earnings

increases (Balachandran and Mohanram 2011).

Table 3 presents statistics underlying figure 1. The first column in panel A presents mean

value relevance for each accounting amount across all sample years. It reveals net income, NI, is

the most relevant accounting amount. NI’s mean VR of 46.5% indicates it provides nearly half of

the combined value relevance of the accounting amounts. However, consistent with the

declining trend in relevance of NI, panel A also reveals this proportion declines almost steadily

from 61.4% in the 1960s to 33.6% in the 2010s. The only decade with an increase is the 2000s,

19 Figure 1 reveals the two years for which the industry indicators have the highest VR are 1979, the year of the

second oil crisis, and 1999, the height of the technology bubble. Also, untabulated findings reveal in 1979 (1999)

Fama-French industry 4, oil and gas, (industries 5 and 6, high technology and telecommunication) accounts for

68.1% (93.1%) of the combined value relevance of the industry indicators. See also footnote 3.

23

which reflects some recovery after the technology bubble. Panel A also reveals equity book

value, BVE, is the second most relevant accounting amount across all sample years (mean VR =

15.4%). However, its value relevance increases noticeably in the 1980s, from 6.3% to 20.6%20

and remains about the after that, except for the 2010s when it declined from 21.6% to 14.8%.

Operating cash flow, CF, exhibits the largest increase in relevance in recent years—from 3.7% in

the 1990s, to 9.4% in the 2000s, to 15.9% in the 2010s when it replaces equity book value as the

second most relevant accounting amount.

Table 3, panel B, presents estimates of 𝛽1 from equation (3). As expected, panel B reveals

the relevance of NI (BVE) significantly declines (increases) (t-stats. = –8.67 and 4.32). Also as

expected, panel B reveals accounting amounts related to intangible assets, growth opportunities,

and alternative performance measures become more relevant. In particular, RD, INTAN, CASH,

REVGR, CF, REV, and SPI all significantly increase in relevance (t-stats. range from 1.68 to

9.40). Only ADV and OCI exhibit no significant trend (t-stats. = 0.38 and –0.37). Consistent

with these individual amounts’ trends, the statistics related to trends in combined relevance for

accounting amounts related to intangible assets, Intans, growth opportunities, Growth, and

20 Untabulated statistics reveal this increase in BVE’s relevance largely occurred in the early 1980s and is associated

with two factors. First, economy-wide earnings were lower during the 1981 and 1982 recession, which likely

resulted in increased relevance of BVE (Barth et al. 1998a; Collins et al. 1999). Relatedly, the economy proportion

of Non-New Economy Loss firms increased and, as figure 2, panel C, reveals, BVE is the most relevant accounting

amount for these firms. Second, the proportion of New Economy firms increased dramatically in the early 1980s.

As figure 2, panel A, reveals, BVE has substantial relevance for these firms. The statistics also reveal the increase in

New Economy firms in the early 1980s is attributable to all types of New Economy firms. Although the increase is

largest for young and loss firms, which is consistent with loss firms explaining BVE’s increased relevance, it is

second largest for technology and young firms, followed by technology and loss and young, technology, and loss

firms. Another possible explanation for BVE’s increased value relevance in the early 1980s is changes in accounting

standards. From 1980 to 1983 approximately forty Statements of Financial Accounting Standards (SFAS) became

effective. However, there was no single major standard. Rather, most of the new SFASs incorporated into the

FASB literature industry-specific guidance previously issued by the AICPA. Regardless, the large number of

standards precludes us from isolating their effects from other changes in the economy.

24

alternative performance measures, AltPerf, reveal each category significantly increases in

relevance (t-stats. = 3.01, 4.00, and 12.26).

Regarding the other accounting amounts, panel B reveals, as expected, a significant

decline in the DIV’s relevance (t-stat. = –4.65).21 Although we have no expectations, panel B

reveals CAPX’s relevance significantly increases (t-stat. = 1.87), COGS’s and ASSETS’s

significantly decline (t-stats. = –2.72 and –2.54), and SGA’s exhibits no significant trend (t-stat.

= 1.63).

Table 3, panel C, presents estimates of 𝛽1 from equation (4) and reveals a significant

increase in the number of important accounting amounts, regardless of threshold (t-stats. range

from 5.45 to 8.36). This finding indicates that over time price reflects information in more

accounting amounts, which reveals a more nuanced relation between the amounts and price.

5.3 Value relevance of accounting amounts for new economy and non-new economy firms

Table 4 presents descriptive statistics for New Economy, Non-New Economy Profit, and

Non-New Economy Loss firms. Panel A presents the proportion of firms comprising each group

and panel B presents distributional statistics, by group, for the variables we use. Panel A reveals

that New Economy, Non-New Economy Profit, and Non-New Economy Loss firms comprise, on

average, 22.1%, 67.4%, and 10.5% of the sample. As expected, panel A also reveals that the

proportion of New Economy and Non-New Economy Loss firms increases significantly at rates

of 0.431 and 0.279 percentage points per year (t-stats. = 4.27 and 9.82), and the proportion of

Non-New Economy Profit firms significantly decreases at a rate of –0.709 percentage points per

year (t-stats. = –6.23). Not surprisingly, panel B reveals mean NI is negative, –0.56 and –1.03,

21 More recently, firms repurchase shares as a substitute for dividends. Share repurchases data are available only

from 1973. See footnote 18. Nonetheless, untabulated findings based on including share repurchases in equation

(1) from 1973 to 2014 reveal a significantly positive relevance trend in that its untabulated equation (3) coefficient

(t-statistic) is 0.026 (5.99).

25

for New Economy and Non-New Economy Loss firms, and positive, 1.62, for Non-New

Economy Profit firms.

Figure 2 shows the evolution in value relevance, VR, of each accounting amount for the

three groups of firms that largely are consistent with our expectations and reveal the changes in

relevance are not limited to New Economy firms.22 Regarding New Economy firms, panel A

shows a steep decline in relevance of NI and a large increase in relevance of other accounting

amounts. As with the full sample in figure 1, the pie charts reveal that in the 1970s the relevance

of NI dominated that of the six other important accounting amounts. Unlike the full sample, two

of these, RD and REV, relate to intangible assets and alternative performance measures. In the

2010s there are ten important accounting amounts, five of which relate to intangible assets,

growth opportunities, and alternative performance measures. Strikingly, NI does not appear in

the 2010s pie chart, which indicates its mean VR is less than 10%.23

Regarding Non-New Economy Profit firms, panel B also shows a decline in NI’s

relevance and an increase in the number of relevant accounting amounts, although—as

expected—neither trend is as pronounced as it is for New Economy firms. However, as with

New Economy firms, amounts related to intangible assets, growth opportunities, and alternative

performance measures are barely noticeable in the 1970s, but explain a large portion of equity

value in the 2010s. As in panel A, the pie charts reveal that in the 1970s NI’s relevance was

dominant, and RD and REV are two of the six other important amounts. In the 2010s NI is less

22 Untabulated findings reveal that the mean combined OOS R2s across years for New Economy, Non-New-

Economy Profit, and Non-New Economy Loss firms are 54.7%, 68.5%, and 51.0%. As expected, each is lower than

71.5% for the full sample over the same period. See footnote 15. 23 Figure 2 reveals the industry indicators have the highest VR in 1988. Untabulated statistics reveal that in 1988

Fama-French industry 6, telecommunications, accounts for 86.7% of the ten industry indicators’ combined value

relevance. Untabulated statistics also reveal telecommunications industry firms experienced a decrease in

profitability in 1988, whereas other firms experienced an increase; from 1987 to 1988 mean NI decreased

(increased) by 0.13 (0.12) in the telecommunications industry (full sample).

26

dominant, and of the nine other important amounts, five relate to intangible assets, growth

opportunities, and alternative performance measures.

Regarding Non-New Economy Loss firms, panel C shows NI has little relevance in any

year. As with New Economy and Non-New Economy Loss firms, amounts related to intangible

assets, growth opportunities, and alternative performance measures were barely noticeable in

1970, but explain a large portion of equity value in more recent years. The pie charts reveal that

in the 1970s, the relevance of BVE dominated that of the other six important accounting

amounts, one of which relates to growth opportunities. In the 2010s BVE is less dominant and of

the eight other important amounts, four relate to intangible assets, growth opportunities, and

alternative performance measures. NI is not important in either decade.

Table 5 presents statistics underlying figure 2. Table 5, panel A, reveals that NI has the

highest mean value relevance across all sample years for New Economy firms and Non-New

Economy Profit firms, 33.3% and 48.4%. However, it has declined for both groups of firms,

although, as expected, more dramatically for New Economy firms—from 73.9% in the 1970s to

only 1.2% in the 2010s for New Economy firms and from 55.6% to 38.8% for Non-New

Economy Profit firms. NI has little relevance for Non-New Economy Loss firms—across all

sample years, in the 1970s, and in the 2010s, mean value relevance was only 0.5%, 0.5%, and

0.3%. For these firms, BVE consistently has the highest mean value relevance, 42.4%, 43.0%,

and 26.9%. The increasing economy proportion of New Economy and Non-New Economy Loss

firms documented in table 4, panel A, and the low relevance of earnings for these firms help

explain the decline in relevance of earnings for the full sample in table 3.

Regarding other accounting amounts, across all sample years, table 5, panel A, reveals

CASH and REVGR, which relate to growth opportunities, are more relevant for New Economy

27

firms than for Non-New Economy firms, as expected. Mean VR for CASH and REVGR are 9.1%

and 5.6% for New Economy firms versus 1.7% and 1.6% (3.6% and 1.2%) for Non-New

Economy Profit (Loss) firms. By the 2010s, CASH and REVGR are among the most relevant

accounting amounts for New Economy firms (20.5% and 12.3%). Also in the 2010s, RD and

INTAN, which relate to intangible assets, as expected, are more relevant for New Economy firms

(14.0% and 2.8%) than for Non-New Economy firms (4.7% and 2.3% for Profit; 2.0% and 2.3%

for Loss). In contrast, DIV is less relevant for New Economy firms than for Non-New Economy

firms (0.8% versus 4.0% for Profit and 5.4% for Loss). The increasing proportion of New

Economy firms documented in table 4, panel A, and the high (low) relevance of CASH, REVGR,

RD, and INTAN (DIV) for these firms contribute to the increase (decline) in relevance for these

accounting amounts in the full sample in table 3.

Table 5, panel B, presents estimates of 𝛽1 from equation (3) that largely are consistent

with our expectations. Panel B reveals NI exhibits a significant relevance decline for New

Economy firms, which is substantially larger than the decline for Non-New Economy Profit

firms (trends = –1.814 and –0.404; t-stats. = –12.02 and –3.98). Although NI exhibits no

significant decline in relevance for Non-New Economy Loss firms (trend = –0.003; t-stat. = –

0.29), recall from panel A that earnings has little relevance for these firms. BVE exhibits a

significant increase in relevance for New Economy firms, which is somewhat larger than the

increase for Non-New Economy Profit firms (trends = 0.223 and 0.170; t-stats. = 1.86 and 2.34).

BVE exhibits a significant relevance decline for Non-New Economy Loss firms (trend = –0.510;

t-stat. = –2.46), but remains the most relevant amount for these firms.

Panel B also reveals accounting amounts related to intangible assets—RD and INTAN—

and growth opportunities—CASH and REVGR—have substantially more positive trends for New

28

Economy firms than for Non-New Economy firms.24 Unexpectedly, but consistent with

economy-wide changes, the trends in accounting amounts related to alternative performance

measures—CF, REV, SPI, and OCI—for New Economy firms are not notably different from

those for other firms.

Panel B reveals significant increases in relevance for INTAN, CASH, and CF for all three

groups of firms. It also reveals that relevance of amounts related to intangible assets, growth

opportunities, and alternative performance measures significantly increase for all groups (t-stats.

range from 2.26 to 12.06), except for amounts related to intangible assets for Non-New Economy

Profit firms, which also increases, but not significantly so (t-stat. = 1.66). These findings suggest

firms emblematic of the new economy are not solely responsible for the increases in relevance of

these accounting amounts. Rather, the increases reflect economy-wide changes.

6. Additional analyses

6.1 Constant sample

We base our main findings on samples that change composition over time. We do this

because we are interested in how value relevance evolves as the economy evolves, which

includes entrance and exit of firms with potentially different business models. However, using a

changing sample precludes us from identifying how relevance of accounting information evolved

for the same set of firms. Thus, we estimate value relevance for the 323 firms that survive from

1971 to 2014. A disadvantage to using this constant sample is that requiring firms to exist for so

many years affects the extent to which our findings apply more generally because such firms had

24 The RD trend is higher for New Economy firms, 0.123, than for Non-New Economy Profit firms and Non-New

Economy Loss firms, 0.012 and 0.049. However, the New Economy firms trend is not significant (t-stat. = 1.57),

which partially reflects the changing composition of New Economy firms. Untabulated statistics reveal the mean

and median age of New Economy firms (full sample) is 5.0 and 3.0 (11.1 and 8.0) years, and the average time a firm

is a New Economy (full sample) firm is 3.7 (10.5) years.

29

sustained, profitable performance. Regardless, the most notable untabulated statistics indicate

earnings is the most relevant accounting amount (mean VR = 57.6%) and its relevance does not

decline (t-stat. = 1.18). In addition, although relevance of amounts related to alternative

performance measures increases significantly (t-stat. = 3.89), those related to intangible assets

and growth opportunities do not (t-stats. = –0.86 and –0.62), which is unsurprising for these

long-established firms.

6.2 Financial firms

We exclude financial firms from our primary sample because their valuations and

operations are different from other firms and accounting research often excludes them or studies

them separately. Nonetheless, we estimate value relevance for financial firms from 1971 to

2014. Untabulated findings reveal, as with the other samples, relevance of NI (BVE) declines

(increases). The findings also reveal OCI and DIV are more relevant for financial firms than for

the other samples, and DIV does not significantly decline in relevance (t-stat. = –1.18).25

Although relevance of accounting amounts related to growth opportunities does not increase

significantly (t-stat. = 0.35), those related to intangible assets and alternative performance

measures do (t-stats. = 6.01 and 4.12), which supports our inference that the increase these

amounts’ relevance is not solely attributable to New Economy firms.26

6.3 Scale effects

As explained in section 3.1, we estimate equation (1) using share-deflated variables to

facilitate comparison with prior research examining trends in value relevance of accounting

25 The DIV finding is consistent with using the dividend discount model to value banks (Damodaran 2009) and with

banks continuing to use dividends to signal future cash flows (Floyd et al. 2015). The OCI finding is consistent with

Dhaliwal et al. (1999), which finds comprehensive income is more value relevant for financial firms. 26 Perhaps surprisingly, RD increases in relevance (t-stat. = 4.31). For financial firms RD can arise from financial

technology and ownership of portfolio firms engaging in research and development.

30

information. However, Brown et al. (1999) expresses concern about using R2 to measure value

relevance in the presence of scale effects. Brown et al. (1999) assumes a multiplicative scale

effect and recommends scaling all variables, except industry indicators, by beginning-of-year

price. More recently, Barth and Clinch (2009) finds deflation by shares generally is effective in

mitigating a variety of scale effects, including multiplicative, in regressions similar to equation

(1), but deflating by beginning-of-year price is not. In addition, deflating by beginning-of-year

price effectively transforms the dependent variable from price to return. Gu (2005) shows the

Brown et al. (1999) inferences relating to changes in R2 over time are attributable to price and

return models differing as to the economic relations they represent rather than to scale effects.

Nonetheless, we estimate equation (1) using OLS with earnings and equity book value as

the only explanatory variables, and scaling by beginning-of-year price. Consistent with Brown

et al. (1999), the untabulated findings reveal a significant decline in value relevance (t-stat. = –

2.63). However, when we estimate a beginning-of-year-price-deflated specification based on

CART using all accounting amounts, untabulated findings reveal no significant decline (t-stat. =

–0.04). These findings suggest the additional accounting amounts, nonlinearities, and

interactions offset the decline in combined relevance of earnings and equity book value.

As an alternative, Brown et al. (1999) recommends including in equation (2) the cross-

sectional coefficients of variation of equity book value and price. We are reluctant to rely on

inferences from this approach because including the coefficient of variation of equity book value

as an explanatory variable likely removes the effect we seek to test.27 Regardless, untabulated

27 The intuition is as follows. Assume price is a linear function of equity book value and random noise. The R2 of a

regression of price on equity book value would increase in the variance of equity book value. Thus, if this variance

increases because equity book value conveys more information about price and nothing else changes, R2 would

increase. However, this also would result in equity book value’s coefficient of variation being a significant

explanatory variable in equation (2) because its numerator (denominator) increases (remains constant). It also

would not be surprising if the time trend coefficient were insignificantly different from zero incremental to this

31

findings based on implementing this approach using CART and all accounting amounts reveal no

significant decline in combined value relevance (t-stat. = –0.47).

7. Conclusion

The question we address is how the value relevance of accounting information has

evolved as the economy transitioned from primarily industrial to a new economy based on

services and information technology. Prior research concludes the value relevance of accounting

information—particularly earnings—has declined, and largely attributes the decline to the shift

to the new economy.

We assess value relevance by determining the power of accounting amounts to explain

equity price. In addition to earnings and equity book value, we consider accounting amounts that

reflect information about intangible assets—research and development expense, recognized

intangible assets, and advertising expense—amounts that reflect information about growth

opportunities—cash and revenue growth—and amounts related to alternative performance

measures—operating cash flow, revenue, special items, other comprehensive income—all of

which we expect are more relevant in the new economy, as well as dividends, capital

expenditures, COGS, SGA, total assets, and ten industry indicators. Prior research predicts and

finds these amounts are value relevant incremental to earnings. However, not all of the amounts

are included in prior research studying trends in value relevance of accounting information.

We base our inferences on estimating the relation between price and accounting amounts

using a flexible, non-parametric approach—Classification and Regression Trees (CART)—

which automatically incorporates nonlinearities and interactions that minimize out-of-sample

prediction errors. We estimate annual relations from 1962 to 2014 and use out-of-sample

variable, even though the increase in equity book value’s coefficient of variation is attributable to the increase in

equity book value’s value relevance. We thank Greg Clinch for insights into this analysis.

32

explanatory power as the value relevance metric. We show CART avoids overstating value

relevance that can occur when using OLS estimation and avoids understating value relevance

because its flexible approach enables the underlying relevance of the accounting amounts to

manifest more fully.

In contrast to prior research, we find no decline in combined value relevance of

accounting information. If anything, we find some evidence of an increase. Consistent with

prior research, we find earnings (equity book value) has become significantly less (more)

relevant. More importantly for our research question, we find accounting amounts related to

intangible assets, growth opportunities, and alternative performance measures become

significantly more relevant. Although we find these trends are most pronounced for firms

emblematic of the new economy, they also are evident for the full sample as well as non-new

economy profit and non-new economy loss firms. These findings indicate that new economy

firms are not solely responsible for the relevance increases. Rather, the increases reflect

economy-wide changes. We also find more accounting amounts have become relevant, which

reveals the relation between price and accounting amounts has become more nuanced.

Our findings provide insights into how accounting information reflects information

investors use in valuing firms in the new economy. The economy-wide increasing relevance of

accounting amounts related to intangible assets, growth opportunities, and alternative

performance measures reveals these categories of information are relevant to investors in the

new economy, despite their incomplete reflection in accounting. This, in turn, suggests

enhancing their reflection in accounting, for example, by improving accounting for intangible

assets, ensuring financial statements include key inputs to investors’ assessments of growth

opportunities, and relaxing the emphasis on a single performance measure, could increase the

33

relevance of accounting information. Our findings also provide insights potentially relevant to

equity valuation. In particular, our findings suggest an increasing number of accounting

amounts, most notably those related to intangible assets, growth opportunities, and alternative

performance measures, reflect information relevant to valuing firms. This trend also is

economy-wide, and not limited to firms emblematic of the new economy.

Taken together, our findings reveal an evolution in value relevance of accounting

information beyond earnings and equity book value to a more nuanced relation between

accounting amounts and share price that reflects the rise in the new economy, not a declining

relation. Thus, our findings indicate it is premature to conclude accounting information has lost

its relevance.

34

References

Aboody, D. M.E. Barth, and R. Kasznik. 2004. SFAS No. 123 stock-based compensation

expense and equity market values. The Accounting Review 79: 251-275.

Aboody, D., and B. Lev. 1998. The value relevance of intangibles: The case of software

capitalization. Journal of Accounting Research 36 (Supplement): 161-191.

Balachandran, S., and P. Mohanram. 2011. Is the decline in value relevance of accounting driven

by increased conservatism? Review of Accounting Studies 16: 272-301.

Barth, M.E., Beaver, W.H., Hand, J.R.M., and W.R. Landsman. 1999. Accruals, cash flows, and

equity values. Review of Accounting Studies 3: 205-229.

Barth, M.E., Beaver, W.H., and W.R. Landsman. 1998a. Relative valuation roles of equity book

value and net income as a function of financial health. Journal of Accounting and

Economics 25: 1-34.

Barth, M.E., Beaver, W.H., and W.R. Landsman. 2001. The relevance of the value relevance

literature for financial accounting standard setting: Another view. Journal of Accounting

and Economics 31: 77-104.

Barth, M.E., Clement, M.B., Foster, G., and R. Kasznik. 1998b. Brand values and capital market

valuation. Review of Accounting Studies 3: 41-68.

Barth, M.E., and G. Clinch. 2009. Scale effects in capital markets-based accounting research.

Journal of Business Finance & Accounting 36(3&4): 253-288.

Barth, M.E., Konchitchki, Y., and W.R. Landsman. 2013. Cost of capital and earnings

transparency. Journal of Accounting and Economics 55: 206-224.

Beaver, W. H., Cascino, S., Correia, M., and M. F. McNichols. 2017. Group Affiliation and

Default Prediction. Available at

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2850458

Bhattacharya, N., Black, E. L., Christensen, T. E., and Larson, C. R. 2003. Assessing the relative

informativeness and permanence of pro forma earnings and GAAP operating earnings.

Journal of Accounting and Economics 36(1): 285-319.

Bradshaw, M. T., and Sloan, R. G. 2002. GAAP versus the street: An empirical assessment of

two alternative definitions of earnings. Journal of Accounting Research 40(1): 41-66.

Breiman, L. 2001. Random Forests. Machine Learning 45(1): 5-32.

35

Brown, S., Lo, K., and T. Lys. 1999. Use of R2 in accounting research: Measuring changes in

value relevance over the last four decades. Journal of Accounting and Economics 28: 83-

115.

Brunnermeier, K., and S. Nagel. 2004. Hedge funds and the technology bubble. Journal of

Finance 59(5): 2013-2040.

Buera, F.J., and J.P. Kaboski. 2012. The rise of the service economy. American Economic

Review 102(6): 2540-2569.

Callen, J.L., Robb, S.W.G., and D. Segal. 2008. Revenue manipulation and restatements by loss

firms. Auditing: A Journal of Practice and Theory 27(2): 1-29.

Collins, D.W., Maydew, E. L., and I. S. Weiss. 1997. Changes in the value-relevance of earnings

and book values over the past forty years. Journal of Accounting and Economics 24: 39-67.

Collins, D. W., Pincus, M., and H. Xie. 1999. Equity valuation and negative earnings: The role

of book value of equity. The Accounting Review 74(1): 29-61.

Core, J. E., Guay, W. R., and A. Van Buskirk. 2003. Market valuations in the New Economy: An

investigation of what has changed. Journal of Accounting and Economics 34: 43-67.

Correia, M., Kang, J., and S. Richardson. 2017. Asset Volatility. Available at:

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2361686.

Damodaran, A. 2009. Valuing financial services firms. Retrieved January 2, 2016 from

http://people.stern.nyu.edu/adamodar/pdfiles/papers/finfirm09.pdf.

Darrough, M., and J. Ye. 2007. Valuation of loss firms in a knowledge-based economy. Review

of Accounting Studies 12: 61-93.

Davern, M., N. Gyles, D. Hanlon, and M. Pinnuck. 2018. Is financial reporting still an effective

tool for equity investors in Australia? Working paper. University of Melbourne.

Davis, A. K. 2002. The value relevance of revenue for internet firms: Does reporting grossed-up

or barter revenue make a difference? Journal of Accounting Research 40(2): 445-477.

Donelson, D. C., Jennings, R., and J. McInnis. 2011. Changes over time in the revenue-expense

relation: Accounting or economics? The Accounting Review 86(3): 945-974.

Dhaliwal, D., Subramanyam, K. R., and R. Trezevant. 1999. Is comprehensive income superior

to net income as a measure of firm performance? Journal of Accounting and Economics 26:

43-67.

http://people.stern.nyu.edu/adamodar/pdfiles/papers/finfirm09.pdf

36

Dichev, I.A., and V.W. Tang. 2008. Matching and the changing properties of accounting

earnings over the last 40 years. The Accounting Review 83(6): 1425-1460.

Dontoh, A., Radhakrishnan, S., and J. Ronen. 2004. The declining value-relevance of accounting

information and non-information-based trading: An empirical analysis. Contemporary

Accounting Research 21(4): 795-812.

Ertimur, Y., Livnat, J., and M. Martikainen. 2003. Differential market reactions to revenue and

expense surprises. Review of Accounting Studies 8: 185-211.

Fama, E.F., and K.R. French. 1992. The cross-section of expected returns. Journal of Finance

47(2): 427-465.

Fama, E.F., and K.R. French. 1993. Common risk factors in the returns on stocks and bonds.

Journal of Financial Economics 33: 3-56.

Fama, E.F., and K.R. French. 2015. A five-factor asset pricing model. Journal of Financial

Economics 116: 1-22.

Faulkender, M., and R. Wang. 2006. Corporate financial policy and the value of cash. Journal of

Finance 61(4): 1957-1990.

Filip, A., A. Ghio, and L. Paugam. 2018. Accounting information in innovative Small and

Medium Entities. Working paper. ESSEC Business School.

Floyd, E., Li, N., and D.J. Skinner. 2015. Payout policy through the financial crisis: The growth

of repurchases and the resilience of dividends. Journal of Financial Economics 118(2):

299-316.

Francis, J., and K. Schipper. 1999. Have financial statements lost their value relevance? Journal

of Accounting Research 37(2): 319-352.

Gerakos, J. J., Hahn, R., Kovrijnykh, A., and F. Zhou. 2016. Prediction versus inducement and

the informational efficiency of going concern opinions. Available at

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2727771.

Gu, F. 2005. Innovation, future earnings, and market efficiency. Journal of Accounting, Auditing

& Finance 20(4): 385-418.

Hastie, T., Tibshirani, R., and J. Friedman. 2001. The Elements of Statistical Learning, Springer,

New York Inc., New York, NY, U.S.A.

Hayn, C. 1995. The information content of losses. Journal of Accounting and Economics 20:

125-153.

37

Holthausen, R.W., and R.L. Watts. 2001. The relevance of the value relevance literature for

financial accounting standard setting. Journal of Accounting and Economics 31: 3-75.

Hribar, P., N.T. Jenkins, and W.B. Johnson. 2006. Stock repurchases as an earnings management

device. Journal of Accounting and Economics 41: 3-27.

Jones, S. 2017. Corporate bankruptcy prediction: a high dimensional analysis. Review of

Accounting Studies 22: 1366-1422.

Jones, D.A., and K. J. Smith. 2011. Comparing the value relevance, predictive value, and

persistence of other comprehensive income and special items. The Accounting Review

86(6): 2047-2073.

Joos, P., and G.A. Plesko. 2005. Valuing loss firms. The Accounting Review 80(3): 847-870.

Lev, B., and F. Gu. 2016. The End of Accounting and the Path Forward for Investors and

Managers. John Wiley & Sons, Inc. Hoboken, NJ, USA.

Lev, B., and T. Sougiannis. 1996. The capitalization, amortization, and value-relevance